Proc mixed

[jescam05]

Hi !

I want to run the next procedure,

data TMT;
input trat rep c1 c2 c3 c4 c5 c6 c7 c8 c9;
y=c1; conteo=1; output;
y=c2; conteo=2; output;
y=c3; conteo=3; output;
y=c4; conteo=4; output;
y=c5; conteo=5; output;
y=c6; conteo=6; output;
y=c7; conteo=7; output;
y=c8; conteo=8; output;
y=c9; conteo=9; output;
drop c1-c9;
datalines;
1	1	57.9	71.4	87.5	74.3	95.6	92.0	94.1	85.7	94.5
1	2	42.9	69.2	90.5	80.0	98.0	96.7	95.9	94.9	93.6
1	3	37.5	71.4	88.9	54.5	100.0	93.8	88.9	100.0	87.5
1	4	30.8	92.3	85.2	82.8	92.5	94.9	88.4	95.5	92.7
2	1	76.0	47.1	80.6	64.7	97.4	85.4	89.5	86.8	91.4
2	2	55.6	31.6	82.4	78.9	96.4	96.8	76.5	90.0	85.7
2	3	33.3	100.0	100.0	80.0	100.0	100.0	94.4	90.5	88.1
2	4	100.0	100.0	100.0	100.0	94.4	100.0	100.0	100.0	75.0
3	1	66.7	3.3	66.7	100.0	94.4	95.2	83.3	94.1	80.6
3	2	75.0	18.2	100.0	93.5	83.0	94.6	82.8	91.2	88.5
3	3	66.7	54.5	84.6	94.4	73.9	95.0	90.5	90.9	95.2
3	4	44.4	25.0	50.0	100.0	100.0	100.0	90.0	100.0	71.4
4	1	57.9	65.0	73.9	52.4	87.0	100.0	74.2	91.7	81.0
4	2	42.9	50.0	50.0	83.3	100.0	100.0	87.5	100.0	100.0
4	3	100.0	100.0	100.0	100.0	100.0	100.0	100.0	100.0	100.0
4	4	50.0	70.4	94.1	56.4	100.0	95.1	84.4	79.1	92.5
;
proc mixed data=TMT;
class trat conteo;
model y = trat conteo trat*conteo;
repeated conteo / type= un sub= rep r rcorr;
random rep; 
lsmeans trat conteo trat*CONTEO/ pdiff adjust=tukey ;
run;

and in the log appears this legend: WARNING: Unable to make hessian positive definite.

What I do wrong? 

[sld · Accepted solution]

You cannot use 

random rep;

in combination with type=un because UN is defined with G matrix =0. See https://faculty.washington.edu/heagerty/Courses/VA-longitudinal/private/Littell-StatMed2000.pdf. If you omit this statement, your model will converge.

Some additional observations:

(1) You probably can find a covariance type that is more parsimonious than UN.

(2) You should use the DDFM option on the MODEL statement; KR2 is usually a good choice.

(3) Your response variable is a percentage and as such assumptions of normality and homogeneity of variance are not well met. Consider either a transformation http://support.sas.com/documentation/cdl/en/stsug/62259/HTML/default/viewer.htm#ugvartransform_sect6...

or the beta distribution in the GLIMMIX procedure.

I hope this helps.

[jescam05]

Excuse me! 

Do you have the procedure to check the assumptions?